In spite of the rapid development of general large language model (LLM) technology, there are significant issues such as hallucination and lack of professionalism in its application. In this case, the large model's ability to meet the specialized Q & A needs of a construction project is limited. Compared with the direct improvement of the internal architecture of LLM, retrieval-augmented generation (RAG) can provide more professional and accurate information support from external knowledge bases when generating answers in LLM. Characterized by simple operation, flexible updating, and strong universality, RAG has become one of the key technologies to enhance the performance of large model professional Q & A. A set of augmented LLM for construction engineering management professional knowledge was constructed based on the RAG and evaluated by the real questions of the licensing examination for construction engineers and registered architects. The results show that the Q & A score rate of RAG large language model is improved by 17.67%, compared with the Qwen-plus general large model. On this basis, RAG large language model Q & A enhancement effect and its limitations are discussed, aiming to promote the in-depth application of RAG technology in construction engineering projects, and to help create AI assistants proficient in the specialized knowledge of construction engineering projects.